Intention to Treat (ITT)

Intention to treat (ITT) is an analysis strategy in clinical trials that includes all randomized participants in the groups to which they were originally assigned, regardless of whether they adhered to the treatment protocol, completed the study, or received the intended intervention. It preserves the benefits of randomization and, when outcomes are available for everyone randomized, gives an unbiased estimate of the effect of assigning the treatment, which reflects its effect in practice to the extent that adherence in the trial resembles routine care.

ITT analysis is the standard primary analysis for superiority randomized controlled trials: ICH E9, the statistical guideline adopted by the FDA and EMA, bases the primary analysis of superiority trials on the intention to treat principle, and the CONSORT 2025 statement asks authors to state who is included in each analysis and in which group (item 21b). Its importance stems from a simple but powerful principle: randomization creates comparable groups at baseline, and any deviation from the original group assignment can reintroduce the confounding that randomization was designed to eliminate. Excluding participants who dropped out, switched treatments, or were nonadherent may bias results because these behaviors are often related to the treatment itself or the outcome of interest. ITT analysis provides a pragmatic estimate of the treatment effect under real world conditions, where imperfect adherence is the norm.

In a strict ITT analysis, every participant who was randomized is included in the analysis, analyzed according to their original group assignment, and counted exactly once. If a participant assigned to the drug group never takes the drug, they remain in the drug group for analysis. If a participant assigned to placebo crosses over to active treatment, they remain in the placebo group. This approach answers the question: "What is the effect of assigning patients to this treatment?" rather than "What is the effect of actually receiving this treatment?" ICH E9(R1), the 2019 addendum on estimands, links the ITT principle to this treatment policy question and asks trials to state in advance which treatment effect (estimand) they target, including how events such as stopping or switching treatment are handled.

Implementing a full ITT analysis requires collecting outcome data on all randomized participants, including those who withdraw from treatment. This is often challenging, as participants who discontinue treatment may also become lost to follow up. Strategies for maximizing outcome ascertainment include collecting data through registries, administrative databases, or brief follow up contacts even for participants who stop treatment.

When outcome data are missing despite best efforts, the ITT principle is preserved by using all available data and appropriate methods for handling missingness, such as multiple imputation or mixed models for repeated measures (MMRM), which rest on assumptions about the missing values that the data cannot confirm and so call for sensitivity analyses. Some trials define a "modified ITT" population that excludes participants who were randomized but never received any treatment or never provided any post baseline data. While common, modified ITT analyses should be clearly distinguished from full ITT analyses and accompanied by sensitivity analyses.

In a superiority trial, the ITT estimate is often conservative because including nonadherent participants dilutes the treatment effect. In noninferiority and equivalence trials the same dilution makes the arms look more similar, so ITT is not necessarily conservative there, and ICH E9 recommends planning both an ITT and a per protocol analysis. A per protocol analysis, restricted to participants who followed the protocol, is often reported alongside ITT as a sensitivity analysis, but it does not simply estimate the effect of receiving the treatment: excluding participants after randomization can bias it in either direction, because those who adhere may differ from those who do not. The two analyses therefore do not reliably bracket the true effect; when the question is the effect of actually receiving treatment, ICH E9(R1) asks trials to define that estimand in advance and choose methods suited to it.

ITT analysis should be the primary analysis in superiority randomized controlled trials, as ICH E9 recommends, while noninferiority and equivalence trials should plan both ITT and per protocol analyses. It is particularly important in pragmatic trials that aim to inform real world treatment decisions, where the question of interest is the effect of the treatment policy (offering the treatment) rather than the biological effect of the treatment itself. In explanatory trials focused on biological mechanisms, per protocol analysis may be of greater interest, but ITT should still be reported. ITT analysis in its strict sense does not apply to observational studies, where participants are not randomized to treatment groups.

A common error is labeling an analysis as ITT when participants with missing outcome data have been excluded, effectively converting it into a complete case analysis. True ITT requires accounting for all randomized participants. Another mistake is interpreting a nonsignificant ITT result as evidence that the treatment does not work, when the result may reflect dilution from nonadherence rather than lack of efficacy. Conversely, some researchers report only the per protocol analysis when the ITT analysis is nonsignificant, which introduces bias and is considered selective reporting. Finally, failing to plan for ITT analysis during the study design phase can result in insufficient efforts to collect outcome data from participants who discontinue treatment.

In a 12 month RCT comparing a new antihypertensive drug to placebo in 500 patients (250 per group), 45 participants in the drug group discontinue treatment due to side effects, and 20 participants in the placebo group are lost to follow up. The ITT analysis includes all 500 randomized participants in their assigned groups: the 45 who stopped the drug are still followed up and measured at 12 months, and outcomes for the 20 lost to follow up are multiply imputed. The mean reduction in systolic blood pressure is 8.2 mmHg in the drug group versus 3.1 mmHg in the placebo group (difference: 5.1 mmHg; with a standard deviation of about 15 mmHg in each group, p < 0.001). The per protocol analysis, restricted to the 435 participants who completed the study per protocol, shows a larger difference (6.8 mmHg, p < 0.001), consistent with dilution of the effect in the ITT population, although the per protocol estimate may also be biased, because the patients who stopped the drug for side effects may differ from those who continued it.